关键词: X光图像, 违禁品检测, 空间和通道重构, 多分支轻量化卷积, 损失函数

Abstract: To resolve the problems of high overlap, occlusion and complex background interference in X-ray contraband images, an improved X-ray contraband detection model based on multi-branch lightweight convolution and attention mechanism for contraband detection is proposed. In the backbone, spatial and channel reconstruction attention mechanism (SCAM) is designed to strengthen the extraction of key features and suppress background interference, and also improve the ability of the model to detect contraband in complex scenes. Using the multi-branch lightweight convolution (MLCB) to reduce the number of model parameters, so as to improve the operation efficiency. In addition, the minimum point distance intersection over union (MPDIoU) loss function and soft non-maximum suppression (Soft NMS) are integrated to replace the complete intersection over union (CIoU) loss function to improve the problems of false positive and false negative caused by overlap of contraband. The proposed model is verified on OPIXray, HIXray and SIXray datasets, and mAP50 reaches 95.7%, 83.7% and 95.3%, respectively. The results show that the proposed method has high accuracy and strong robustness, and it can solve the problems of overlap and occlusion effectively.

Key words: X-ray images, contraband detection, Spatial and Channel reconstruction Attention, Multi-branch Light Convolution, Loss function